hubert-base-960h-itw-deepfake

This model is a fine-tuned version of facebook/hubert-base-ls960 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.0756
Accuracy: 0.9873
FAR: 0.0083
FRR: 0.0203
EER: 0.0143

Model description

Quick Use

  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

  config = AutoConfig.from_pretrained("abhishtagatya/hubert-base-960h-itw-deepfake")
  feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained("abhishtagatya/hubert-base-960h-itw-deepfake")

  model = HubertForSequenceClassification.from_pretrained("abhishtagatya/hubert-base-960h-itw-deepfake", config=config).to(device)

  # Your Logic Here

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-06
train_batch_size: 2
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	FAR	FRR	EER
0.4081	0.39	2500	0.1152	0.9722	0.0285	0.0267	0.0276
0.1168	0.79	5000	0.0822	0.9844	0.0120	0.0216	0.0168
0.0979	1.18	7500	0.0896	0.9846	0.0130	0.0195	0.0162
0.0983	1.57	10000	0.1007	0.9833	0.0155	0.0186	0.0171
0.0901	1.97	12500	0.0756	0.9873	0.0083	0.0203	0.0143

Framework versions

Transformers 4.38.0.dev0
Pytorch 2.1.2+cu121
Datasets 2.16.2.dev0
Tokenizers 0.15.1

abhishtagatya
/

hubert-base-960h-itw-deepfake